Single-Image Rain Removal Network Based on an Attention Mechanism and a Residual Structure

Image rain removal involves eliminating the impact of rain on an image to increase its visual quality. In this paper, we propose a single-image rain removal network based on an attention mechanism and a residual structure. The proposed method extracts high-level semantic information and low-level detail information via a feature aggregation module and a residual channel attention mechanism, respectively, and fuses the two feature parts through a fusion tail to learn a mapping to the rain removal image. The attention mechanism unequally processes different features and pixels to improve the generalization of the image deraining procedure. A continuous residual structure is used to prevent problems such as overfitting and gradient disappearance during network training. In an ablation study, we show that both the feature aggregation module and the residual channel attention mechanism significantly improve the resulting image quality in terms of the PSNR and SSIM index measure. Moreover, the obtained experimental results demonstrate that our method quantitatively and qualitatively outperforms the current state-of-the-art approaches on the Rain100L, Rain100H and real-world image datasets.